Run voice cloning and video dubbing locally without an internet connection
You can clone your voice, dub videos with subtitles, and convert long texts into audiobooks on a laptop or desktop without an internet connection. VoiceStudio is an open-source tool that handles these voice generation and conversion processes using models installed on your device. It supports 646 languages and allows you to download only the models you need, giving you direct control over storage space and performance.
Voice workflows in a local environment
The core of this tool is that all processing happens in a local environment. Because data is not sent to cloud servers, it is suitable for tasks where privacy is important. In addition to the default engine, OmniVoice, you can select dozens of other engines from the model manager. Only the model files required for each engine are downloaded, so unused engines do not take up space. It runs on CPU without a GPU, though speed is slower, and it supports NVIDIA CUDA, AMD ROCm, and Apple Metal for hardware acceleration.
Four ways to handle voice
Users generate or transform voice through four main features.
- Voice cloning Upload a clean reference recording to have new text read in that voice. It generates new content while maintaining the characteristics of the original speaker.
- Video dubbing Upload a video to translate the dialogue and generate a dubbed video that matches the timing of the original audio. Subtitles and audio are synchronized naturally.
- Voice design Describe the desired voice characteristics in text to generate a new voice that matches them. You can create a voice with a specific tone even without an existing recording.
- Speech recognition and reading Use a floating widget to input text via voice or convert long documents into audiobook format. Batch processing allows you to work on multiple files at once.
Installation and platform support
It supports a wide range of platforms, including macOS, Windows, Linux, and mobile devices. macOS supports Metal acceleration on Apple Silicon (M1 or later), while Intel Macs require a remote backend connection. Windows and Linux support NVIDIA CUDA and AMD ROCm, and Linux can also run GGUF engines via Vulkan. Android and iOS are provided as preview versions and work offline after model download. Installation methods vary by platform, including Homebrew, PowerShell scripts, Docker, or downloading APK/IPA files.
Things to check before use
This tool follows the AGPL-3.0 license, which allows commercial use but requires compliance with its conditions. Model weights and generated outputs have separate licenses, so you must check the license of each model in the model manager. In particular, the voice cloning feature cannot be used without the consent of the voice owner. In environments without a GPU, it runs on CPU only, which can significantly slow down processing speed, so you should consider hardware specifications for large-scale tasks. Additionally, some models take up significant space, so it is advisable to check if you have sufficient storage beforehand.
